inference oom

[WuJian1995]

when inference a100 with 80GB memory:

torch.cuda.OutOfMemoryError: CUDA out of memory. Tried to allocate 57.93 GiB. GPU 0 has a total capacty of 79.35 GiB of which 17.27 GiB is free. Process 41187 has 62.08 GiB memory in use. Of the allocated memory 60.67 GiB is allocated by PyTorch, and 11.17 MiB is reserved by PyTorch but unallocated. If reserved but unallocated memory is large try setting max_split_size_mb to avoid fragmentation.  See documentation for Memory Management and PYTORCH_CUDA_ALLOC_CONF

how to solve it ?

[zhangtianshu]

You may not install a suitable environment. One A100 with 80G is enough for inference on TableLlama.